Object-based stereoscopic conversion of MPEG-4 encoded data

초록

Stereoscopic conversion of two-dimensional (2-D) video is considered in object-based approach that independently processes each video object. Our works extend the previous frame-based stereoscopic conversion of MPEG-1 and 2 to MPEG-4. In MPEG-4, each image is composed of a background object and primary object(s). In the first step, a camera motion type is determined for generating a stereoscopic background image. For this, motion vectors of a background object are utilized. The generation of a stereoscopic background object makes use of a current image and a previous image. As well, The stereoscopic primary object uses a current image and its horizontally-shifted version to avoid the possible vertical parallax that might happen. In the second step.. the two stereoscopic objects are combined to generate a stereoscopic image. As verified in experiments performed on two MPEG-4 sequences, the object-based stereoscopic conversion can cope with a vertical parallax that has been a difficult problem to deal with in the frame-based approach.